diff --git a/sumpy/expansion/local.py b/sumpy/expansion/local.py
index 9e891a2dc48399bfded9e7cb0fc6ed63f38c40e4..483bc0eb16fcee7b43d54aedb4eaa8092f593620 100644
--- a/sumpy/expansion/local.py
+++ b/sumpy/expansion/local.py
@@ -130,7 +130,7 @@ class VolumeTaylorLocalExpansionBase(LocalExpansionBase):
         evaluated_coeffs = (
             self.derivative_wrangler.get_full_kernel_derivatives_from_stored(
                 coeffs, rscale))
-        bvec = bvec / rscale
+        bvec = bvec * rscale**-1
         result = sum(
                 coeff
                 * mi_power(bvec, mi)
@@ -195,7 +195,7 @@ class VolumeTaylorLocalExpansionBase(LocalExpansionBase):
             # canceling "rscales" closer to each other in the hope of helping
             # with that.
             result = [
-                    (taker.diff(mi) * tgt_rscale**sum(mi)).expand(deep=False)
+                    (taker.diff(mi) * tgt_rscale**sum(mi)).expand()
                     for mi in self.get_coefficient_identifiers()]
 
         logger.info("building translation operator: done")
diff --git a/sumpy/expansion/multipole.py b/sumpy/expansion/multipole.py
index 4db9250b329602ead272e122c0c128059829277d..0ddbb428108ff0bd52dd896f88ba1115a66428c5 100644
--- a/sumpy/expansion/multipole.py
+++ b/sumpy/expansion/multipole.py
@@ -88,7 +88,7 @@ class VolumeTaylorMultipoleExpansionBase(MultipoleExpansionBase):
             for i, mi in enumerate(coeff_identifiers):
                 result[i] /= (mi_factorial(mi) * rscale ** sum(mi))
         else:
-            avec = avec/rscale
+            avec = avec * rscale**-1
 
             result = [
                     mi_power(avec, mi) / mi_factorial(mi)
@@ -107,7 +107,7 @@ class VolumeTaylorMultipoleExpansionBase(MultipoleExpansionBase):
                     self.kernel.adjust_for_kernel_scaling(
                         vector_subs(
                             expr,
-                            bvec, bvec/rscale),
+                            bvec, bvec * rscale**-1),
                         rscale, nderivatives)
                     / rscale ** (nderivatives - nderivatives_for_scaling))
         else:
diff --git a/sumpy/symbolic.py b/sumpy/symbolic.py
index 55f91284d62c536c290c2ccf8f2c07f4d829796f..665aff9d80b6fb9f5ae0e67770a361110db36210 100644
--- a/sumpy/symbolic.py
+++ b/sumpy/symbolic.py
@@ -199,7 +199,7 @@ def vector_subs(expr, from_vec, to_vec):
         for icol in range(from_vec.cols):
             subs_pairs.append((from_vec[irow, icol], to_vec[irow, icol]))
 
-    return expr.subs(subs_pairs)
+    return expr.subs(dict(subs_pairs))
 
 
 def find_power_of(base, prod):
diff --git a/test/test_kernels.py b/test/test_kernels.py
index 65295e29451e039619ef65076926cc1b5b182d22..a0cf8ae7f34a529d60e1cc07d45af87c854d3db8 100644
--- a/test/test_kernels.py
+++ b/test/test_kernels.py
@@ -141,6 +141,9 @@ def test_p2e2p(ctx_getter, base_knl, expn_class, order, with_source_derivative):
     from sympy.core.cache import clear_cache
     clear_cache()
 
+    from sumpy import set_caching_enabled
+    set_caching_enabled(False)
+
     ctx = ctx_getter()
     queue = cl.CommandQueue(ctx)
 
@@ -331,6 +334,8 @@ def test_p2e2p(ctx_getter, base_knl, expn_class, order, with_source_derivative):
             slack += 0.3
             grad_slack += 0.3
 
+    set_caching_enabled(True)
+
     assert eoc_rec_pot.order_estimate() > tgt_order - slack
     assert eoc_rec_grad_x.order_estimate() > tgt_order_grad - grad_slack
 
@@ -350,6 +355,9 @@ def test_translations(ctx_getter, knl, local_expn_class, mpole_expn_class):
     from sympy.core.cache import clear_cache
     clear_cache()
 
+    from sumpy import set_caching_enabled
+    set_caching_enabled(False)
+
     ctx = ctx_getter()
     queue = cl.CommandQueue(ctx)
 
@@ -597,6 +605,8 @@ def test_translations(ctx_getter, knl, local_expn_class, mpole_expn_class):
 
         pconv_verifier_full.add_data_point(order, err)
 
+    set_caching_enabled(True)
+
     for name, verifier in [
             ("p2m2p", pconv_verifier_p2m2p),
             ("p2m2m2p", pconv_verifier_p2m2m2p),